EMG Pattern Classification Using Neural Networks

The functioning of electromyogram (EMG) driven prosthesis to control the performance of artificial prosthetic arms placed on people with missing limbs depends on the cumulative effect of multiple dynamic factors, some of which include electrode placement position, muscle contraction levels, forearm orientations, etc. However, the study of the combined influence of these dynamic factors has been limited and hence offered us scope to improve the accuracy of the previous studies. We used the data to extract multiple features through the Time Dependent Power Spectrum Descriptor (TD-PSD) algorithm, which has proven to be one of the best methods of feature extraction. Samples are classified using the Neural Pattern Recognition Toolbox with scaled conjugate gradient backpropagation as the training algorithm, which gives an improved accuracy over Support Vector Machine (SVM) classifier. Neural Network is trained using the EMG signals of 10 subjects performing multiple hand movements to achieve classification accuracy up to 94.7%. The results obtained are a testimony to the fact that the suggested method is competent to improve the operation of pattern recognition myoelectric signals.